Development/Plasticity/Repair

Precise Lamination of Retinal Axons Generates MultipleParallel Input Pathways in the Tectum

Estuardo Robles,1,2 Alessandro Filosa,1,2 and Herwig Baier1,2

1Max Planck Institute of Neurobiology, D-82152 Martinsried, Germany, and 2Department of Physiology, University of California, San Francisco, SanFrancisco, California 94158

The axons of retinal ganglion cells (RGCs) form topographic connections in the optic tectum, recreating a two-dimensional map of thevisual field in the midbrain. RGC axons are also targeted to specific positions along the laminar axis of the tectum. Understanding thesensory transformations performed by the tectum requires identification of the rules that control the formation of synaptic laminae byRGC axons. However, there is little information regarding the spatial relationships between multiple axons as they establish laminar andretinotopic arborization fields within the same region of neuropil. Moreover, the contribution of RGC axon lamination to the processingof visual information is unknown. We used Brainbow genetic labeling to visualize groups of individually identifiable axons during theassembly of a precise laminar map in the larval zebrafish tectum. Live imaging of multiple RGCs revealed that axons target specificsublaminar positions during initial innervation and maintain their relative laminar positions throughout early larval development,ruling out a model for lamina selection based on iterative refinements. During this period of laminar stability, RGC arbors undergostructural rearrangements that shift their relative retinotopic positions. Analysis of cell-type-specific lamination patterns revealed thatdistinct combinations of RGCs converge to form each sublamina, and this input heterogeneity correlates with different functionalresponses to visual stimuli. These findings suggest that lamina-specific sorting of retinal inputs provides an anatomical blueprint for theintegration of visual features in the tectum.

IntroductionRetinal ganglion cells (RGCs) with diverse morphologies and re-sponse properties transmit a retinal representation of the sensoryworld to the brain. In zebrafish, the vast majority of RGCs inner-vate the optic tectum, which initiates behaviorally relevant motorprograms in response to visual cues (Nevin et al., 2010). Withinthe tectal neuropil, axonal arbors are targeted along the retino-topic axes, as well as the laminar axis. Although there is evidenceimplicating guidance cues in the establishment of each map(McLaughlin and O’Leary, 2005; Luo and Flanagan, 2007; Xiao etal., 2011), basic organizing principles of the retinotectal pro-jection remain poorly understood. Previous studies in the lar-val zebrafish demonstrated that RGC axons target specifictectal laminae (Xiao and Baier, 2007). However, several as-pects of lamina assembly have not been directly examinedbecause of the fact that previous studies could not monitor thelamination of multiple, individually identifiable axons withinthe same volume of neuropil.

Recent studies in the mammalian visual system have demon-strated a correlation between cell type and axon lamination pat-tern (Huberman et al., 2008, 2009; Kim et al., 2010; Hong et al.,2011), suggesting that laminar targeting is one determinant ofsynaptic specificity. Cell-type-specific synapse formation mayprovide a mechanism by which diverse retinal inputs are inte-grated to confer complex visual response properties to collicularneurons (Wang et al., 2010). Early anatomical studies demon-strated that RGC axon lamination in the fish tectum is moreprecise than that in the mammalian superior colliculus (SC;Ramon y Cajal, 1995), suggesting that lamination plays a moreprominent role in determining synapse specificity in the fish tec-tum. More recently, in vivo imaging in larval zebrafish has re-vealed that RGCs form planar arbors in the tectum (Xiao andBaier, 2007; Xiao et al., 2011). However, it is unknown whetherRGC-type-specific lamination directs the formation of spatiallysegregated tectal circuits dedicated to processing different typesof visual input.

We used multicolor genetic labeling to monitor the precisionand dynamics of lamina assembly in the developing zebrafishtectum. We characterized a fine-grained sublaminar map gener-ated by precise costratification of retinal axons. Live imaging ofmultiple axons in the same tectum confirmed that laminar posi-tion is established during initial innervation. Multi-day time-lapse imaging demonstrated that relative laminar positions arefixed throughout early larval development, ruling out a process inwhich axonal translocation along the laminar axis serves to refinepatterns of connectivity. Whereas relative laminar positions arestable, retinal arbors in every lamina invariably shift their relative

Received Oct. 24, 2012; revised Jan. 16, 2013; accepted Jan. 30, 2013.Author contributions: E.R., A.F., and H.B. designed research; E.R. and A.F. performed research; E.R. and A.F.

analyzed data; E.R., A.F., and H.B. wrote the paper.This work was supported by the Max Planck Society, the National Science Foundation, and Ford Foundation

postdoctoral fellowships (E.R.), an EMBO long-term fellowship (A.F.), and National Institutes of Health GrantsEY12406 and EY13855 (H.B.). Experiments were initiated at the University of California, San Francisco and concludedat the Max Planck Institute of Neurobiology. We are grateful to members of the Baier laboratory for critical readingof this manuscript and Joe Donovan for assistance with Monte Carlo simulation.

Correspondence should be addressed to Estuardo Robles, Max Planck Institute of Neurobiology, Genes–Circuits–Behavior, Am Klopferspitz 18, D-82152 Martinsried, Germany. E-mail: erobles@neuro.mpg.de.

DOI:10.1523/JNEUROSCI.4990-12.2013Copyright © 2013 the authors 0270-6474/13/335027-13$15.00/0

The Journal of Neuroscience, March 13, 2013 • 33(11):5027–5039 • 5027

retinotopic positions. Examining the relationship between RGCtype and axon lamination pattern in the tectum revealed thateach sublamina is innervated by multiple RGC types and, in mostcases, distinct combinations of RGCs. Functional imaging of visu-ally evoked calcium responses in RGC axons confirmed that differ-ences in axon composition of sublaminae correlate with differentfunctional properties. Our findings support a model in which preciselaminar organization of retinal afferents creates a structural frame-work for the integration of visual inputs to the tectum.

Materials and MethodsFish breeding. We raised zebrafish at 28°C on a 14/10 h light/dark cycle.The Tg(hsp70l:Cre)zdf13 fish line was generously provided by LeonardZon (Harvard University, Cambridge, MA). All procedures were ap-proved by the Committee on Animal Research of the University of Cal-ifornia, San Francisco.

Plasmid construction and transgenesis. The UAS:Brb1.0L transgenesisplasmid was constructed by excising a 14xUAS:Brb1.0L DNA fragmentfrom the pFrank:cytoBrainbow plasmid (a gift from A. Pan and A. Schier,Harvard University, Cambridge, MA) and ligating it into the pBH (Bleed-ing Heart) Tol2 plasmid (a gift from M. Nonet, Washington University,St. Louis, MO). This plasmid contains a transgenesis marker (cmlc2:mCherry) that drives heart-specific expression of mCherry. Transgenesiswas performed by injecting the Tol2 UAS:Brb1.0L plasmid along withtransposase RNA synthesized from the pCS2TP plasmid (a gift from K.Kawakami, National Institute of Genetics, Mishima, Japan). Tg(UAS:Brb1.0L)s1997t fish were identified by mCherry heart expression at 3 dpostfertilization (dpf). Founders with multiple copies were identifiedby determining the percentage of offspring containing the transgen-esis marker. For all Brainbow experiments, only Tg(UAS:Brb1.0L)s1997t fish that produced �90% transgene-containingoffspring were used, ensuring that these fish more than likely containedat least three copies of the Brainbow expression cassette. Transient mo-saic Brainbow expression in RGCs was achieved by injection of the Isl2b:GAL4VP16 plasmid (Campbell et al., 2007) (a gift from C.-B. Chien,University of Utah, Salt Lake City, UT), along with I-Sce I DNA endonu-clease into Tg(UAS:Brb1.0L)s1997t;Tg(hsp70l:cre)zdf13 embryos. AllDNA constructs were pressure injected at a concentration of 25–50 ng/�linto one- to four-cell-stage embryos.

Image acquisition. For all experiments, larvae were treated with 0.003%1-phenyl-2-thiocarbamide starting at 22 h after fertilization to inhibitformation of pigment in skin cells. Larvae were imaged on a fluorescentstereoscope to identify larvae with the appropriate labeling density and asuitable level of recombination. Selected embryos were anesthetized in0.016% tricaine dissolved in E3 medium and embedded in 2% low-melting-point agarose dissolved in E3 medium. Larvae used for multiday imagingwere released from the agarose after each imaging session and placed in aPetri dish filled with E3 medium. Imaging was performed on Carl ZeissPascal and LSM-780 confocal microscopes equipped with a multiline argonlaser for excitation of Cerulean (458 nm) and enhanced yellow fluorescentprotein (EYFP) (514 nm) and a green helium–neon laser to excite dTomato(543 nm). Optical sections were acquired using 1 �m z-steps.

Image analyses. Image stacks were visualized and analyzed using Im-ageJ FIJI software (http://fiji.sc/Fiji). The 3D Viewer plugin written by B.Schmid (Wurzburg University, Wurzburg, Germany) was used to makevolume renderings to remove skin autofluorescence from image vol-umes. Because of the curvature of the tectum, removal of these regionswas necessary to visualize the tectal neuropil from top and side views.Manual tracings and brightness/contrast enhancement were performedin Adobe Photoshop (Adobe Systems). Arbor morphology analysis andtracing of convex polygons used criteria established by Smear et al.(2007) and were performed in FIJI. For measurements of arbor distancefrom the surface of the tectal neuropil, we used side views in which theskin autofluorescence was not removed so that it could be used as amarker for the skin. All data are presented as mean � SEM.

Immunohistochemistry. Anesthetized larvae were fixed overnight in asolution of 4% paraformaldehyde in PBS. Fifty micrometer vibratomesections were cut from larvae embedded in gelatin/albumin and stained

with a chick anti-GFP primary antibody (GeneTex) and an Alexa Fluor-488 anti-chick secondary antibody (Invitrogen).

Two-photon Ca2� imaging and data analyses. For imaging, wecrossed Tg(Atoh7:Gal4-VP16)s1992t to Tg(UAS:GCaMP3)zf350 into themitfa�/� (nacre) background (Del Bene et al., 2010). Larvae were em-bedded in 2% low-melting-point agarose at 6 dpf and imaged at 7 dpfwith a custom-built two-photon microscope equipped with a mode-locked titanium:sapphire Chameleon UltraII laser (Coherent) tuned to920 nm and controlled by ScanImage version 3.6 software (Pologruto etal., 2003). Image time series were acquired with a 40� water-immersionobjective (numerical aperture 0.8; Olympus) at 3.37 Hz. Visual stimulus(a vertical gray bar on a black background, �24° high and 1° wide,moving horizontally back and forth for 2 s for a distance of �8°) wasgenerated with VisionEgg software (Straw, 2008) and presented with an800 � 600 pixel organic light-emitting diode (eMagin) with a green lightfilter (Rosco) to the left eye of the larvae. Imaging was performed in theright (contralateral) tectal neuropil. If x–y motion was present, imagetime series were x–y motion corrected with a program written in MAT-LAB (a gift from D. Tank, Princeton University, Princeton, NJ; modifiedby C. Niell, University of Oregon, Eugene OR; Dombeck et al., 2007).Data with z motion were discarded. Image analysis was performed withImageJ FIJI software. First, �F image time series were generated, and thenregions of interest (ROIs) were detected semiautomatically by thresholding�F image series and automatic detection of thresholded elements using the“analyze particles” function in ImageJ FIJI. �F/F plots were created withMicrosoft Excel software using fluorescence intensity data obtained from theoriginal image time series with NIH ImageJ. These plots were used to classifyROIs as ON, OFF, and ON–OFF events. To assign sublaminar position intwo-photon image series, which were acquired at a single z position, thedistance between the skin and the lower border of stratum fibrosum et gri-seum superficiale (SFGS) was divided into eight sublayers. Two-wayANOVA was performed using MATLAB.

Monte Carlo simulation. Monte Carlo simulation for exact inferencewas performed using StatXact software developed by Cytel. The data inFigure 7K were treated as a contingency table in which sublaminar posi-tion was treated as an ordered variable as a result of the fact that sublami-nae are naturally ordered from superficial to deeper positions, whereasRGC type was treated as a nominal variable. The 10 6 Monte Carlo sam-ples were used to obtain an exact two-sided p value using a Kruskal–Wallis test.

ResultsA transgenic Brainbow system for multicolor labeling inlarval zebrafishBrainbow multicolor fluorescent labeling has been used to studyneural networks and cell lineages in the mouse nervous systemusing stable transgenics containing Brainbow expression cas-settes under the control of cell-type-specific promoter elements(Livet et al., 2007; Card et al., 2011; Pan et al., 2011). In theDrosophila nervous system, stable transgenic lines containingBrainbow expression cassettes under the control of the upstreamactivating sequence (UAS) permit labeling of neural subsetswhen used in conjunction with tissue-specific Gal4 transgenics(Hadjieconomou et al., 2011; Hampel et al., 2011). For multi-color labeling in zebrafish larvae, we used a version of the UAS:Brainbow 1.0 L (UAS:Brb1.0L) cassette that encodes threecytoplasmic fluorescent proteins (dTomato, Cerulean, andEYFP; Fig. 1A) (Pan et al., 2011). To confirm that this Brainbowcassette can generate diverse color combinations, we coinjectedUAS:Brb1.0L and RNA encoding Cre recombinase into embryoscarrying Gal4s1013t. This transgenic line labels several brain re-gions, as well as axial muscle (Scott and Baier, 2009). In someembryos, labeled muscle cells exhibited a variety of color labels(Fig. 1A), confirming that random Cre-mediated recombinationcan generate sufficient color diversity. However, in most injectedembryos, there was a low degree of color diversity, likely attrib-

5028 • J. Neurosci., March 13, 2013 • 33(11):5027–5039 Robles et al. • Sublaminar Targeting of Retinal Axons

utable to early Cre expression inducing recombination in earlymuscle progenitors. This is consistent with other reports thathave found the need to temporally control Cre expression(Weber et al., 2009; Pan et al., 2011).

The retinotectal projection forms multiple sublaminaeTo examine the larval retinotectal projection using Brainbow la-beling, we developed a system to mosaically label RGCs withsufficient color diversity to distinguish between arbors in close

proximity. The generation of diverse color labels in Brainbowtransgenics is contingent on multiple insertions into the genome(Livet et al., 2007). We generated a stable transgenic line contain-ing three or more copies of the UAS:Brb1.0L cassette (Pan et al.,2011). These fish were crossed to Tg(hsp70l:Cre)zdf13, a trans-genic line expressing Cre recombinase under control of the heat-shock-inducible promoter hsp70l (Le et al., 2007). Embryosgenerated from this cross were injected with a DNA plasmid driv-ing Gal4 expression under control of the Islet2b promoter, a

Figure 1. Multicolor Brainbow labeling in transgenic zebrafish. A, Side view of a 3 dpf Gal4s1013t embryo coinjected with DNA encoding UAS:Brainbow and RNA encoding Cre recombinase.Numbered squares are subregions of corresponding muscle cells. Rostral is to the left. B, Top view of the tectum from a 7 dpf UAS:Brainbow, hsp70:Cre larva injected with islet2b:gal4 DNA to drivemosaic Brainbow expression in a subset of RGCs. C, Side view of tectum in B. Image volume was three-dimensionally rendered and rotated to visualize the laminar organization of the retinotectalprojection in the tectum. D, Red– green– blue (RGB) single-channel (middle) and grayscale (right) fluorescence intensity plots of region indicated by yellow box in the left. Black numbered linesindicate distinct fluorescence intensity peaks, whereas gray lines indicate regions of similar thickness without fluorescence peaks. Scale bars: A–C, 20 �m.

Robles et al. • Sublaminar Targeting of Retinal Axons J. Neurosci., March 13, 2013 • 33(11):5027–5039 • 5029

marker for RGCs (Pittman et al., 2008). Early experiments re-vealed that Cre-induced recombination of the Brainbow expres-sion cassette occurred in the absence of heat shock treatment,suggesting “leaky” expression from the hsp70 promoter. How-ever, the random nature of Cre expression resulted in subsets oflarvae with diverse color labels in retinal axons (Fig. 1B). Injec-tion of Islet2b:gal4 DNA yielded mosaic labeling of RGCs andgenerated specimens with either dense (�12 RGCs) or sparse(�12 RGCs) labeling of the retinotectal projection. In larval ze-brafish, the tectal neuropil is oriented at an �45° angle to thedorsoventral axis, located beneath an autofluorescent layer ofskin cells at the tectal surface. To examine the retinotectal projec-tion in confocal image volumes, we used a 3D “clipping” tool toisolate the tectal neuropil (see Material and Methods). Excisedvolumes were examined in the “top-view” orientation to examinearbors along the retinotopic axes and rotated to examine axonsalong the laminar axis.

The tectal neuropil of zebrafish contains four retinorecipientzones: the stratum opticum (SO), SFGS, stratum griseum cen-trale (SGC), and a layer situated at the boundary between thestratum album centrale and the stratum periventriculare (SAC/SPV) (Meek, 1983; Xiao et al., 2005). The majority of retinalinputs innervate the SO and SFGS layers. Non-retinorecipientlayers are located on either side of the SGC, separating the SGCfrom the SAC/SPV and the SFGS, respectively. Rotation of con-focal image volumes with dense RGC labeling allowed us to mea-sure the thickness of these retinorecipient layers along thelaminar axis (SO, 8.3 � 0.3 �m; SFGS, 32.1 � 0.8; SGC, 9.8 � 0.6;SAC/SPV, 8.4 � 0.4; n � 9 larvae; Fig. 1C). Detailed examinationof Brainbow specimens with densely labeled neuropil suggestedthat the retinotectal projection is subdivided into multiple sub-laminae (Fig. 1B,C). The total fluorescence intensity plot of thistectum (Fig. 1D, right) supports the presence of discrete layers inthe SGC, SAC/SPV, and multiple positions within the SO andSFGS (indicated by black lines), as well as known retinorecipientregions devoid of labeling in this particular sample (gray lines).Fluorescence intensity plots in individual red (tdTomato),green (EYFP), and blue (Cerulean) color channels demon-strate the difference in fluorescent protein ratios at differentlaminar positions.

The sublaminar map is formed by thin, regularlyspaced arborsTo directly examine whether the entire tectal neuropil could bedivided into thin sublaminae, we analyzed the average thicknessof retinal axon arbors throughout all layers of the tectum (5.2 �0.2 �m, n � 120). Based on the thickness of SFGS and SO, wesubdivided the retinorecipient neuropil layers into 10 sublami-nae: the SGC and SAC/SPV, which are discrete layers bordered bynon-retinorecipient neuropil, two layers in the SO, and six layersin the SFGS. For statistical comparisons, groups of axons at dif-ferent laminar depths were correspondingly binned into 1 of 10sublaminar designations: SO1 and SO2, SFGS1–SFGS6, SGC, orSAC/SPV (see Materials and Methods). The presence of 10 dis-tinct sublaminae is a close approximation, and future studiesusing more precise methods will be required to determine theirexact number in the tectum.

To examine the spacing between axons within these laminarsubdivisions, we examined larvae with sparse labeling (�12 ax-ons), which made it possible to trace the complex morphologiesof multiple arbors in a small volume of tissue. A demonstration ofthis is presented in Figure 2, A and B, in which four axons withoverlapping retinotopic arborization fields could easily be distin-

guished based on their unique color labels: orange, pink, cyan,and teal. These composite color labels were generated by differentratios of the three Brainbow fluorescent proteins, as shown insingle-channel fluorescence intensity plots (Fig. 2C). The side-view rotation and fluorescence intensity plots revealed that axonsin adjacent positions are oriented in parallel and generally exhibitregular spacing along the laminar axis. Analysis of the interpeakdistance between axons assigned to adjacent sublaminae, such asthe two pairs of axons in Figure 2A–C, revealed spacing that wassimilar to the average axon thickness (5.75 � 0.51, n � 16 axonpairs). Morphological analysis of single axons in each sublaminarevealed that the SO1, SO2, and SFGS1–SFGS6 layers containedaxon arbors with average thicknesses between 4 and 6 �m,whereas the SGC and SAC/SPV layers contained axons between 6and 7 �m thick (Fig. 2D). It should be noted that the variabilitywe observe in both arbor thickness and sublaminar spacing sug-gest that the sublaminar map does exhibit a degree of overlap.However, the predominance of thin, regularly spaced arbors sug-gests that a large fraction of axons contribute to a precise sub-laminar scaffold in the tectum.

Sublaminae are formed by precise costratification of axonsIn specimens with strong color diversity, it was possible to tracethe complex morphologies of multiple arbors within a single sub-lamina. The image volume in Figure 3, A and B, presents anexample of sparse labeling in which several sublaminae containlabeled axons and one of these layers, SO1, contains three arbors(yellow, teal, and red). A sublaminar region isolated from thisvolume by 3D clipping revealed the overlapping retinotopic ar-borization fields of these axons (Fig. 3C). A second example,shown in Figure 3, D and E, contains an SFGS sublamina in whichfive different axon regions could be distinguished based on theircolor profile. The sublaminar volume contains four spectrallydistinct axon arbors in orange, cyan, magenta, and blue and asegment of green-labeled axon that arborized in an adjacent sub-lamina (Fig. 3F). The composite color labels generated by Brain-bow were uniform throughout the axon, which allowed us totrace the fine morphology of arbors with the same sublaminarposition and overlapping retinotopic arborization fields (Fig.3G,I). It should be noted that the number or length of branchesmay be an underestimate because of cases in which every segmentof an arbor could not be faithfully traced as a result of dim label-ing (Fig. 3F, blue arbor). However, in samples with bright label-ing and good color separation, we were able to confidently tracethe full morphology of most arbors within a given sublaminarregion (Fig. 3G, yellow and teal arbors; I, blue, cyan, and orangearbors). Precisely aligned groups of arbors were observed in everyretinorecipient layer of the tectum (data not shown), suggestingthat precise layering is formed through axon costratification.

Although the development of retinotectal topography hasbeen thoroughly characterized, there is little information on thedegree of overlap between axon terminals. The ability to visualizegroups of axons within small neuropil volumes allowed us todirectly examine retinotopic overlap. Our identification of dis-crete sublaminae in the tectum suggests that the most informativeoverlap values are those between RGC axons in the same sub-lamina, because these are more likely to contact the same tectalneurons and interact with each other. To quantify retinotopicoverlap, we measured the area of overlap between two arborsrelative to the sum of the total area encompassed by the two arborareas, in which perfect overlap is 100%. For example, the magentaand cyan axons in Figure 3, I and J, exhibited a degree of retino-topic overlap of 26.7%. The maximum value we observed was

5030 • J. Neurosci., March 13, 2013 • 33(11):5027–5039 Robles et al. • Sublaminar Targeting of Retinal Axons

77.1%, and values �40% were observed in several SO and SFGSsublaminae (data not shown and Fig. 6). Although the randomnature of this labeling does not allow us to calculate the averageoverlap area within different layers, the maximum values ob-served suggest that a high degree of arbor overlap is a generalfeature of retinal afferents in the tectum.

RGC axon laminar targeting is precise ab initioRGC arbor morphology in the larval zebrafish is highly dynamicand arises through iterative extension and retraction of axonal

branches (Hua et al., 2005; Meyer andSmith, 2006; Ben Fredj et al., 2010). Thelaminar precision we observe could ariseab initio (from the beginning) or throughgradual refinement of an arbor that is ini-tially only coarsely targeted. To visualizethe process of RGC axon lamination, itwas first necessary to confirm that uniquecolor labels generated by Brainbow expres-sion are stable enough to allow individualneurons to be monitored over time. Formulti-day imaging, sparsely labeled larvaewere immobilized, imaged, and subse-quently returned to embryo-rearing solu-tion. Repeated imaging at 1–2 d intervalsbetween 5 and 12 dpf revealed only subtlechanges in color labels, predominantly as aresult of a decrease in fluorescence intensityof the Cerulean protein that is likely attrib-utable to photobleaching (data not shown).However, in every larva with good initialcolor diversity, the reduction in Ceruleanfluorescence did not impede our ability toidentify the same neurons at each timepoint.

To determine whether RGC laminarposition is achieved through gradual re-finement, we examined RGC laminar pro-files at early (4 – 6 dpf) and late (10 –12dpf) stages of retinal arborization. Themajority of RGC arbors exhibited preciselaminar targeting at the early time point(37 of 40 larvae examined; Fig. 4A–C, blueand teal arbors indicated by arrows) anddid not further refine. Only three arborsimaged at the early time point hadcoarsely targeted arbors �10 �m thick,which likely spanned two or more sub-laminae (Fig. 4G). Two of these arbors un-derwent a subsequent correction thatresulted in a reduced final arbor thickness.In these cases, the refinement arose througha simultaneous increase in arbor densityin one sublamina (Fig. 4D–F, arrowhead)and a decrease in the other. Cases such asthis may represent a small RGC subpopu-lation that is delayed in sublaminar target-ing or correction of rare targeting errors.We also observed one multi-laminar RGCarbor that did not undergo developmentalrefinement (data not shown). However,the rarity of these cases precluded addi-tional examination. In summary, multi-day imaging revealed that, for the vast

majority of RGCs, sublaminar targeting does not arise throughovergrowth and selective pruning but rather through preciselaminar targeting ab initio.

The laminar map is invariant throughout earlylarval developmentMulti-day imaging of sparse Brainbow labeling in the retinotectalprojection revealed the time course by which RGC arbors adopt athin, sublamina-specific morphology. However, it does not prove

Figure 2. The retinotectal projection is organized into thin sublaminae. A, Top view of a 7 dpf larva with mosaic Brainbowlabeling achieved by injection of islet2b:gal4 plasmid into UAS:Brainbow, hsp70:Cre embryos. In this sparsely labeled specimen,four axons with unique color profiles are labeled. B, Side view of image volume in A. Labeled axons were located in two pairs ofadjacent sublaminae: SO1, SO2, SFGS2, and SFGS3. Note that the arbors have relatively uniform thicknesses and are aligned inparallel. C, RGB fluorescence intensity plot along line indicated by arrowheads in B. Arrows indicate the peak fluorescence fororange, magenta, cyan, and teal arbors, respectively. D, Average thickness for axons in each layer. Scale bars: A, B, represent 20�m.

Robles et al. • Sublaminar Targeting of Retinal Axons J. Neurosci., March 13, 2013 • 33(11):5027–5039 • 5031

that the laminar map is fixed throughout early larval develop-ment. Individual axons maintain relatively thin arbors through-out this period but could shift in position relative to one another.To examine this possibility, we observed relative laminar posi-tioning over time in larvae that contained multiple RGC arborswith unique color labels at different sublaminar positions. Asshown in Figure 5A–D, the morphology and relative laminar po-sition of arbors with unique color labels could be faithfully mon-itored between early and late time points. Although axonmorphologies were dynamic, in every larvae analyzed, we foundthat there were only minor changes in the distance between pairsof arbors along the laminar axis (0.88 � 0.12 �m, n � 26 axonpairs).

Although these findings suggest that the laminar map is stableduring larval development, it does not exclude the possibility thatpreexisting sublaminae split to form two or more new layers. If

this were true, we would expect to observe instances in whicharbors initially innervating the same sublamina subsequently seg-regated to different laminar positions. We conducted multi-dayimaging of larvae in which two RGC arbors with unique colorlabels innervated the same sublamina and overlapped in the reti-notopic plane (Fig. 5E,F). Axons targeted to the same sublaminanever segregated during the observation period (Fig. 5G–K; n �8). Together, these findings indicate that, during early larvaldevelopment, the positioning and axonal composition of sub-laminae is invariant, ruling out a role for laminar translocation inaxonal target field selection.

Dynamic changes in arbor morphology are restricted to theretinotopic axesPrevious studies of RGC axons during early larval developmenthave indicated that retinal arbor morphology is highly dynamic.

Figure 3. Precise lamination of multiple axons in the same sublamina. A, B, Top and side view of a tectum containing 10 labeled RGC axons. Note several precisely laminated groups of axons(arrows) present in several sublaminar positions. C, Top view of an excised volume containing only the SO1 region of volume in A and B. Note three axons with overlapping arborization fields anddistinct composite color labels. Arrows indicate several laminated axons at different sublaminar positions. D, E, Top and side views of a tectum containing 10 labeled RGC axons. F, Top view of anexcised volume containing only the SFGS3 region of volume in D and E. The volume contains four spectrally distinct axon arbors labeled in blue, orange, red, and cyan in addition to a segment of agreen labeled axon that arborized in an adjacent lamina (asterisk). G, H, Line tracings of arbors in C and F. Individual tracings (right panels) show the fine morphology of each arbor. I, J, Convexpolygons bounding the arbors in G and J. These areas were used to determine the retinotopic arborization field of single axons. Scale bars, 25 �m.

5032 • J. Neurosci., March 13, 2013 • 33(11):5027–5039 Robles et al. • Sublaminar Targeting of Retinal Axons

For example, as the zebrafish tectum increases in size, shifting ofaxon arbors in the posterior direction may function to preservethe visuotopic map (Xiao and Baier, 2007). There is also evidencesuggesting that RGC axon arborization is influenced by activity-dependent processes (Hua et al., 2005; Smear et al., 2007; BenFredj et al., 2010). Although the evidence for retinotopic struc-tural plasticity is very strong, previous studies did not examinethe positions of multiple axons with respect to their sublaminarposition. Examination of image volumes containing axons withdifferent laminar positions and overlapping retinotopic arboriza-tion fields revealed that relative retinotopic position is highlydynamic (Fig. 6A). We used axonal tracings to determine changesin arbor overlap and in the distance between the centroids of thearea of each arbor. This analysis revealed that alignment along theretinotopic axes could increase or decrease with similar likeli-hood (Fig. 6C,D). To exclude a model in which these changesreflect shifting of entire sublaminae relative to each other, weexamined arbors occupying the same sublaminar position (Fig.6B). Both area of overlap and centroid distance between sucharbor pairs were highly dynamic and could increase or decrease(Fig. 6C,D). The fact that the degree of retinotopic overlap be-tween neighboring axons in the same sublamina often increasedargues against a simple model in which competition betweenretinal axons reduces arborization field overlap.

Multiple RGC types innervate each sublaminaThe sublaminar map we identified could serve to sort axons basedon the type of visual information they convey. Each tectal sub-lamina could serve a unique function by receiving visual infor-mation from a single RGC type. An extreme alternative is one inwhich each RGC type indiscriminately innervates sublaminaeand functionally diverse circuits are established through selectivesynapse formation. Studies in the mouse retinocollicular projec-tion have demonstrated that different types of RGCs have mor-phologically distinguishable axon terminals (Huberman et al.,2008, 2009; Kim et al., 2010; Hong et al., 2011). To further assessaxonal morphology, we analyzed retinotopic arborization areaand branch density of axons within each sublamina (Table 1).

Morphometric analysis revealed that arbors in both SGC andSAC/SPV sublaminae contained arbors with significantly lowerbranch densities than those in SO1, SO2, and SFGS1–SFGS6[one-way ANOVA (F(9,120) � 8.598, p � 0), followed by Tukey–Kramer post hoc test, p 0.05]. However, there were no significantdifferences in laminar thickness [one-way ANOVA (F(9,161) � 2.108,p � 0.032), followed by Tukey–Kramer post hoc test, p � 0.05] orarbor area [one-way ANOVA (F(9,124) � 1.435, p � 0.18), followedby Tukey–Kramer post hoc test, p � 0.05]. We did observe a consid-erable degree of morphological variability within individual sub-laminae, which could reflect innervation by multiple axon types withdistinct morphologies.

To directly determine the RGC types that innervate specificsublaminae, we examined the stratification pattern of RGC den-drites in the retinal inner plexiform layer (IPL), which is a hall-mark feature of RGC type identity. This analysis requiredaldehyde fixation and vibratome sectioning of the larval eyes.Brainbow labeling was not compatible with this treatment be-cause of diminished fluorescence intensity after fixation and theinability to distinguish the different fluorescent proteins by im-munofluorescent staining. Therefore, we genetically labeled sin-gle RGCs via coinjection of Isl2b:Gal4 and UAS:EGFP DNAconstructs into wild-type embryos. Larvae containing a singlelabeled RGC were fixed, sectioned, and immunostained to corre-late axon laminar position with RGC dendrite morphology in theIPL. In the adult zebrafish retina, dendritic morphology has beenused to classify 11 RGC types based on dendritic morphology inthe IPL (Mangrum et al., 2002; Ott et al., 2007). Based on thesepublished nomenclatures, we grouped larval RGC cell types intosix major classes: types 1/5/6, 2/3, 4, 7/8, 10, and 11. These majorclasses exhibit unique dendrite stratification patterns in the IPL,which most likely reflect distinct functional properties attribut-able to synaptic input from different subsets of bipolar and ama-crine cells. The example presented in Figure 7, A and B, shows thetecta of two larvae, each with a single SO1 arbor positioned �2�m from the tectal surface (Fig. 7C,D, arrows). Retrospectiveimmunofluorescent labeling in the retina revealed that these ax-ons originated from RGCs with different dendritic stratification

Figure 4. The majority of RGC axons target specific sublaminae ab initio. A, Side view of an RGB image volume containing three RGC arbors labeled in red, blue, and teal. B, C, Grayscale imagesof green and blue channels in A. Note that both the blue and teal arbors are�5 �m thick in the synaptic plane (arrows). D–F, Grayscale images of red channel at 6, 8, and 10 dpf. Note that the initiallybushy, diffuse arbor gradually becomes more refined through elaboration of the arbor region in the deeper laminar position (arrowhead) and a near-complete loss of branches in other regions. G,Change in arbor thicknesses for 40 RGC axons between early (4 – 6 dpf) and late (10 –12 dpf) stages. Note that the majority of arbors do not significantly refine their laminar thickness. Red arrowindicates data points for axon shown in D–F. Scale bars, 25 �m.

Robles et al. • Sublaminar Targeting of Retinal Axons J. Neurosci., March 13, 2013 • 33(11):5027–5039 • 5033

patterns, one bistratified type 10 RGC(Fig. 7E) and one monostratified type 2/3RGC with a dendrite targeting an OFFlayer of the IPL (Fig. 7F). The examplepresented in Figure 7, G and H, is an im-age volume of a single tectum containingtwo distinct arbors that both innervateSFGS1. One of these axons originatedfrom a monostratified type 4 RGC and theother from a bistratified type 10 RGC (Fig.7 I, J). These findings confirm that morethan one type of RGC can innervate thesame sublamina.

In total, retrospective cell type analysiswas performed on 74 RGCs with axonsinnervating a single tectal sublamina.Overall, we found that 9 of the 10 tectalsublaminae are innervated by multipleRGC types (Fig. 7K), although it shouldbe noted that sample sizes for SGC andSAC/SPV sublaminae were limited be-cause of their sparse innervation. Further-more, although our data indicate thateach RGC class innervates multiple sub-laminae, current estimates suggest thatthe number of RGC types in the vertebrateretina is �20 (Badea et al., 2009; Volgyi etal., 2009). Therefore, each class we definedmost likely comprises multiple types, andwe cannot rule out that single RGC typesdo, in fact, innervate single sublaminae.Our data also identify several sublaminaeinnervated by unique yet overlappingcombinations of RGC types. For example,the SO1, SO2, and SFGS1 sublaminaewere each innervated by two RGC types:type 10 bistratified RGCs and one addi-tional RGC type, 1/5/6, 2/3, and 4, respec-tively. The data presented in Figure 7Ksuggest that differential innervation ofsublaminae by each RGC types follows astereotyped pattern. Alternatively, basedon a merely qualitative inspection of thedata, we could not rule out that the RGCsdestined to project to the tectum makestochastic choices among individual sub-laminae. Therefore, to exclude this possi-bility, we used Monte Carlo simulation toperform exact inference on this categori-cal dataset (Kruskal–Wallis test; for de-tails, see Materials and Methods). Thisstatistic yielded a significant two-sided pvalue of 0.013, which supports the exis-tence of a stereotyped code for sublaminarinnervation by RGC axons.

Sublamina-specific responses tovisual stimulationOur finding that unique combinations of RGC axons innervateeach tectal sublamina suggests that these layers transmit differentvisual information to retinorecipient neurons in the tectum. Totest this hypothesis, we conducted functional imaging of the reti-notectal projection in larvae doubly transgenic for Atoh7:Gal4

and UAS:GCaMP3, which express the calcium indicatorGCaMP3 in a majority of RGCs (Fig. 8A) but not in tectal neu-rons (Del Bene et al., 2010). Preliminary experiments establishedthat a small moving bar near the center of the visual field evokedcalcium responses in the neuropil more reliably than stationary

Figure 5. Laminar position is fixed throughout early larval development. A, B, Top views at 6 and 12 dpf of a tectum containingthree RGC arbors with overlapping retinotopic positions: a magenta arbor, a yellow arbor, and a purple arbor. C, D, Side views of theimage volumes in A and B show laminar position of the three RGC axons. Relative distances between the arbors, as indicated bybrackets, do not change between time points. E, F, Top views at 5 and 10 dpf of an RGB image volume containing two RGC arborsinnervating the same tectal sublamina with overlapping retinotopic positions: a green arbor and a red arbor. G–K, RGB andsingle-channel grayscale side views of image volume in E and F show that the identical laminar position of these RGCs is maintainedfrom 5 to 10 dpf. Scale bar, 20 �m.

5034 • J. Neurosci., March 13, 2013 • 33(11):5027–5039 Robles et al. • Sublaminar Targeting of Retinal Axons

flashing spots. A previous study demon-strated that a large percentage of tectalneurons are responsive to moving stimuli(Niell and Smith, 2005). The stimulusused in our study was a vertical gray bar�24° high and 1° wide moving horizon-tally back and forth across a black back-ground at a speed of 16°/s. Stimuluspresentation typically resulted in tran-sient calcium elevations in multiple re-gions throughout the tectal neuropil (Fig.8B–D). GCaMP3 fluorescence was not de-tectable in SGC or SAC/SPV, possibly be-cause of the sparse innervation of theselayers. Therefore, our analysis was restrictedto the SO and SFGS layers, which containthe majority of retinal inputs (Meek, 1983).To compare between axonal ensembles atdifferent laminar positions, we subdividedthe GCaMP3-labeled neuropil into eight re-gions of equal thickness, which approxi-mate the SO1, SO2, and SFGS1–SFGS6designations used above.

GCaMP3 fluorescence intensity changeswere often detected at stimulus onset and/orstimulus offset, and these temporally sepa-rated responses were used to categorize sub-regions of the neuropil as ON, OFF, or ON–OFF (Fig. 8E,F). Because of the nature ofthe visual stimulus used (moving bar with aleading and a trailing edge), these classifica-tions do not necessarily fit the classical cate-gorization of ON, OFF, and ON–OFF RGCsbut rather are meant to reflect the generalresponses of RGC axons to the appearanceor disappearance of the stimulus. Our anal-ysis revealed that ON, OFF, and ON–OFFresponsive regions were present in everyretinorecipient layer. However, these classesof inputs were not equally distributedamong tectal sublaminae (Fig. 8G). Nota-bly, there were opposing gradients of ONand OFF inputs, with ON responses pre-dominant in superficial sublaminae,whereas OFF responses were predominantin deep sublaminae. ON–OFF inputs werepredominant in intermediate sublaminae.

The combination of these three gradients resulted in each layer hav-ing a distinct functional identity as a result of its particular ratio ofON, OFF, and ON–OFF responsive subregions (F(14,336) � 12.57,p � 0, two-way ANOVA, class � layer interaction). For example, thetwo most superficial layers contained predominantly ON inputs,whereas the two deepest layers contained primarily OFF inputs. Inmultiple instances, directly adjacent layers differed in their relativeratios of ON, OFF, and ON–OFF inputs received. Together,these findings confirm that anatomically defined sublaminaeexhibit different responses to visual stimuli.

DiscussionThis study has revealed that the retinotectal projection of larvalzebrafish is anatomically and functionally divided into fine sub-laminae. Previous studies demonstrated that a subpopulation ofRGCs labeled by a pou4f3 transgene form thin, planar arbors

Figure 6. Retinotopic mapping is dynamic during early larval development. A, Line tracings of the yellow and purplearbors in Figure 5, A and B. Note changes in arbor morphology and increase in spatial overlap. Colored polygons correspondto the area occupied by each arbor, and the colored square within indicates the arbor centroid. Regions of arbor overlap areindicated by black. Note marked increase in arbor overlap and decrease in centroid distance. B, Line tracings of the greenand red arbors in Figure 5, E and F. C, D, Quantification of developmental changes in overlap area and centroid distance foraxon pairs innervating different (interlaminar) or identical (intralaminar) tectal sublaminae. Note that, in each case, bothoverlap and centroid distance are dynamic during larval development.

Table 1. Sublamina-specific morphological analysis of retinal arbors

Tectal sublaminaArbor thickness(�m) Arbor area (�m 2)

Branch density(per 1000 �m 2)

SO1 (n � 18) 4.24 � 0.23 935.09 � 85.14 16.81 � 1.39SO2 (n � 21) 5.76 � 0.49 1255.9 � 88.07 14.55 � 0.82SFGS1 (n � 18) 5.31 � 0.41 1241.84 � 92.75 16.64 � 1.18SFGS2 (n � 15) 5.77 � 0.56 1200.29 � 107.75 17.63 � 1.05SFGS3 (n � 17) 5.84 � 0.39 1281.92 � 154.72 14.94 � 1.30SFGS4 (n � 21) 5.62 � 0.49 1456.6 � 174.25 13.72 � 0.78SFGS5 (n � 20) 5.57 � 0.43 1335.18 � 111.51 14.59 � 0.86SFGS6 (n � 21) 4.73 � 0.31 1251.2 � 130.72 15.36 � 0.96SGC (n � 18) 6.34 � 0.62 1336.96 � 107.13 7.37 � 0.54*SGC (n � 13) 6.68 � 0.61 1038.49 � 126.36 8.03 � 0.56*

Note differences in arbor area and branch density. All data are presented as mean � SEM. One-way ANOVA (F(9,120)

� 8.598, p � 0), followed by Tukey–Kramer post hoc test. *p 0.05.

Robles et al. • Sublaminar Targeting of Retinal Axons J. Neurosci., March 13, 2013 • 33(11):5027–5039 • 5035

during tectal innervation (Xiao and Baier, 2007; Gosse et al.,2008). We substantially extended these findings by demonstrat-ing that lamination serves to spatially segregate inputs from RGCafferents based on the type of information they convey. Groups ofaxons precisely costratify to form individual sublaminae, each atwo-dimensional representation of the visual field. Furthermore,these parallel retinotopic maps are formed by axons originatingfrom multiple different combinations of RGC types. However,our cell-type classification is an underestimate of RGC diversity.Vertebrate retinas contain �20 distinct RGC types (Masland,2001; Mangrum et al., 2002; Rockhill et al., 2002; Badea et al.,2009; Volgyi et al., 2009; Hong et al., 2011). It is possible thatspecific RGC subtypes within these major RGC classes do target asingle sublamina. However, we can exclude that individual sub-laminae receive input from a single RGC type. Future studies will

enable us to further differentiate RGC classes based on additionalcriteria, such as differential expression of molecular markers anddistinct visual response properties.

Lamina-restricted axon terminals have been described in thetectum/SC of several mammalian and avian species (Sachs et al.,1986; Inoue and Sanes, 1997; May, 2006; Huberman et al., 2008,2009; Kim et al., 2010). In the mouse, RGC axons can adoptlamina-restricted projections through either accurate targetingor gradual refinement of an initially broad arbor (Huberman etal., 2008; Kim et al., 2010). Multi-day imaging of individual larvaeallowed us to accurately monitor relative laminar positions invivo during a stage of development characterized by dynamicchanges in the morphology of RGC arbors (5–12 dpf; Hua et al.,2005; Meyer and Smith, 2006; Ben Fredj et al., 2010). Our datareveal that the vast majority of axon arbors do not translocate

Figure 7. Individual sublaminae are innervated by multiple RGC types. A, B, Top views of single GFP-labeled arbors in two different 7 dpf larvae. C, D, Side views of the axons in A and B magnified4�. Proximity to the skin overlying the tectum (autofluorescence in the CFP channel, blue; arrows) was used to identify these as SO1-targeting RGC arbors. E, F, Retrospective GFP immunofluo-rescence in the retinas of larvae in A and B. Boundaries of the IPL are indicated by white lines. Note that, although the RGC in E has a bistratified dendritic arbor, the RGC in F has a monostratified arborin an OFF layer of the IPL. G, H, Top and side views of two GFP-labeled arbors in the same tectum that were manually pseudocolored green and purple. Tectal surface is indicated by white line in H.Note that these arbors have adjacent retinotopic positions and target the same tectal sublamina (SFGS1). I, J, Dendritic morphology of RGCs corresponding to axonal arbors in G and H. Note that,although the RGC in I has a bistratified dendritic arbor, the RGC in J has a monostratified arbor in the center of the IPL. Scale bars: A, B, G, H, 25 �m; C, D, 5 �m; E, F, I, J, 20 �m. K, Table summarizingthe distribution of RGC axons targeting each retinorecipient sublaminae.

5036 • J. Neurosci., March 13, 2013 • 33(11):5027–5039 Robles et al. • Sublaminar Targeting of Retinal Axons

along the laminar axis, thereby maintaining the relative laminarpositions established during initial innervation (Fig. 9). It is sur-prising that the tectal sublaminar map is not further refined dur-ing early larval development given that, during this period, theretinal IPL is expanding and RGC dendrites are increasing incomplexity to target newly formed laminae (Mumm et al., 2006).Given the strong evidence that some types of RGC axons in themouse SC concurrently undergo laminar and retinotopic refine-ment (Simon and O’Leary, 1992; Huberman et al., 2008; Kim etal., 2010), this may represent a fundamental difference in the

lamination process of RGC axons in the zebrafish tectum com-pared with the mouse SC. However, in both systems, RGC axonlamination does not require visually evoked or spontaneous ac-tivity (Huberman et al., 2008; Nevin et al., 2008). Furthermore,our analysis of laminar innervation patterns of different RGCclasses is generally consistent with RGC-type-specific laminationin the mammalian SC described by Hong et al. (2011), althoughadditional work will be required to determine the precise numberof RGC types in the zebrafish and make functional correlationswith identified RGC types in mouse.

Figure 8. Functional characterization of sublaminar targeting by RGC types. A–C, Image showing the fluorescence of GCaMP3 in the tectal neuropil of an Atoh7:Gal4, UAS:GCaMP3 zebrafish larva.The surface of the tectum is to the top. B–D, Two-photon pseudocolored images acquired from a subregion of the tectal neuropil of a 7 dpf during baseline (B), at stimulus onset (“ON,” C), and afterdisappearance (“OFF,” D). E, Distribution of ON, OFF, and ON–OFF RGC axons in the neuropil detected in the experiment shown in B–D. F, Sample traces of ON, OFF, and ON–OFF ROIs taken from thetectum shown in A–D. G, Quantification of percentage ON, OFF, and ON–OFF RGC inputs along the laminar axis. Error bars represent SEM. Numbers in A, E, and G represent approximate sublaminardivisions of the tectal neuropil from most superficial (1) to the deepest (8) layer of the SFGS. Scale bar: A–E, 10 �m.

Figure 9. RGC axonal arborization field selection during larval development. Schematic depicting the process of RGC axon arborization along both the retinotopic and laminar axes. A, Lamina selection occursduring initial growth into the tectum. The axons of RGC types with different dendritic morphologies in the IPL (labeled red, green, and blue) select specific retinotopic and sublaminar positions within theretinorecipient layersofthetectum(SO,SFGS,SGC,SAC/SPV).Notethattheredandgreenaxonscostratify inthemostsuperficialsublaminaofSFGS. B,Subsequently, therelative laminarpositionsbetweentheseaxons is maintained as axons form branched, planar arbors. Concurrently, some tectal neurons extend dendrites that target specific sublaminae (cell in black) also form arbors in specific sublaminae. C, This initialarborization is followed by a period during which relative laminar positions remain unchanged, whereas iterative retinotopic refinements occur in parallel.

Robles et al. • Sublaminar Targeting of Retinal Axons J. Neurosci., March 13, 2013 • 33(11):5027–5039 • 5037

Numerous studies have reported that retinal axon arbors arehighly dynamic during development and that their branchingand growth is influenced by neuronal activity (Hua et al., 2005;Smear et al., 2007; Ben Fredj et al., 2010). In the mammalianretinocollicular projection, axons initially overshoot their appro-priate retinotopic position and subsequently prune excessivebranches in a process that requires spontaneous retinal waves(McLaughlin et al., 2003). Our findings demonstrate that retinalarbors invariably shift their relative retinotopic positions whilemaintaining a fixed laminar position. One mechanism that hasbeen postulated to underlie retinotopic map refinement is com-petition between RGC axons for postsynaptic targets. Althoughthere is evidence that competition between neighboring axonarbors influences retinal arbor size and complexity (Hua et al.,2005; Ben Fredj et al., 2010), previous studies did not examine therelative laminar position of imaged axons. Our data suggest thatthese competitive interactions are most likely to occur amongaxons in the same sublamina. The fact that degree of overlapbetween axon pairs in the same sublamina can be large, and oftenincrease during development, rules out simple models in whichretinal arbors contract to effectively tile the retinorecipient neu-ropil. However, individual sublaminae are innervated by multi-ple RGC types; therefore, we cannot exclude the possibility thataxons of the same type form non-overlapping mosaics.

Our data indicate that the precise layering of the zebrafishretinotectal projection is generated by the costratification of ax-ons from multiple RGC types. This precision is matched by thedendritic morphologies of several identified neuron types in thetectum (Scott and Baier, 2009; Del Bene et al., 2010; Robles et al.,2011; Gabriel et al., 2012). This is in contrast to observations inthe SC of nocturnal rodents, in which both retinal arbors (Honget al., 2011) and collicular neuron dendrites (Mooney et al., 1984;Lee et al., 1997; Endo and Isa, 2001) are more diffusely targeted.In each system, appropriate connections between retinal axonsand retinorecipient neurons are likely formed through targetedlamination in conjunction with selective synapse formation(Huberman et al., 2010). Our data support a model in whichaccurate circuit formation in the visual system of zebrafish reliesmore heavily on laminar precision than that of rodents. Thissuggests that, in fish, the morphology of tectal neuron dendritesmay directly reflect the complement of retinal inputs it receives.For example, we previously characterized a class of bistratifiedperiventricular interneuron (bsPVIN) that forms a preciselystratified dendritic arbor within the most superficial layer of theSFGS (Robles et al., 2011). Based on our present findings, wepredict that inputs from RGCs of classes 4 and 10 specificallyconverge onto bsPVIN dendrites (Fig. 9). However, the tectumalso contains neurons with dendrites spanning multiple sublami-nae (Scott and Baier, 2009; Robles et al., 2011), suggesting thattectal dendrite lamination may reflect the diversity of inputs aneuron receives.

It is well established in various vertebrate species that RGCtypes classified by dendritic morphology have distinct physiolog-ical properties (for review, see Masland, 2001). Calcium imagingof retinal arbors in the tectum confirmed that tectal sublaminaeexhibit divergent responses to a small moving stimulus, whichprovides a functional correlate to their distinct innervation pat-terns. Our findings are consistent with recent studies examiningthe organization of direction- and orientation-selective retinalinputs along the laminar axis of the tectum (Gabriel et al., 2012;Nikolaou et al., 2012). Specifically, these experiments revealedthat the SO and superficial SFGS preferentially receive direction-selective inputs. Our data provide an anatomical basis for these

spatially segregated functional responses, because these layers re-ceive input from bistratified type 10 RGCs, which are morpho-logically similar to direction-selective ganglion cells in mammals(for review, see Wei and Feller, 2011; Vaney et al., 2012). To-gether, these findings reveal parallel functional maps in the tec-tum that are generated by stereotyped patterns of RGC axonlamination. The early formation of these functionally divergentlayers suggests an essential role in the development of tectal neu-ron response properties (Niell and Smith, 2005). Furthermore,the stability of these sublaminae in the presence of tectal neuronswith lamina-restricted dendrites (Robles et al., 2011) suggests amodel in which the type of inputs onto tectal neurons is constant,whereas the retinotopic precision of these inputs is improving.Such refinements can reduce the size of tectal neuron receptivefields and thereby increase visual acuity (Smear et al., 2007). Thisis consonant with the finding that, between 4 and 9 dpf, tectalneurons exhibit a robust increase in responsiveness to small spots(Niell and Smith, 2005). Our findings predict that, during thisdevelopmental window, subpopulations of neurons becometuned to smaller objects without changes in responsiveness toother stimulus features.

The identification of rules governing the spatial organizationof retinal afferents is a necessary step toward understanding howvisual information is processed. Neural maps of sensory input tothe brain can be either continuous or discrete (Luo and Flanagan,2007). The visual pathway harbors both types of maps. The reti-notopic map preserves the spatial order of inputs in the tectumand is therefore continuous, whereas the discrete laminar map weidentified is generated by cell-type-specific lamination.

ReferencesBadea TC, Cahill H, Ecker J, Hattar S, Nathans J (2009) Distinct roles of

transcription factors brn3a and brn3b in controlling the development,morphology, and function of retinal ganglion cells. Neuron 61:852– 864.CrossRef Medline

Ben Fredj N, Hammond S, Otsuna H, Chien CB, Burrone J, Meyer MP(2010) Synaptic activity and activity-dependent competition regulatesaxon arbor maturation, growth arrest, and territory in the retinotectalprojection. J Neurosci 30:10939 –10951. CrossRef Medline

Campbell DS, Stringham SA, Timm A, Xiao T, Law MY, Baier H, Nonet ML,Chien CB (2007) Slit1a inhibits retinal ganglion cell arborization andsynaptogenesis via Robo2-dependent and -independent pathways. Neu-ron 55:231–245. CrossRef Medline

Card JP, Kobiler O, McCambridge J, Ebdlahad S, Shan Z, Raizada MK, SvedAF, Enquist LW (2011) Microdissection of neural networks by condi-tional reporter expression from a Brainbow herpesvirus. Proc Natl AcadSci U S A 108:3377–3382. CrossRef Medline

Del Bene F, Wyart C, Robles E, Tran A, Looger L, Scott EK, Isacoff EY, BaierH (2010) Filtering of visual information in the tectum by an identifiedneural circuit. Science 330:669 – 673. CrossRef Medline

Dombeck DA, Khabbaz AN, Collman F, Adelman TL, Tank DW (2007)Imaging large-scale neural activity with cellular resolution in awake, mo-bile mice. Neuron 56:43–57. CrossRef Medline

Endo T, Isa T (2001) Functionally different AMPA-type glutamate recep-tors in morphologically identified neurons in rat superficial superior col-liculus. Neuroscience 108:129 –141. CrossRef Medline

Gabriel JP, Trivedi CA, Maurer CM, Ryu S, Bollmann JH (2012) Layer-specific targeting of direction-selective neurons in the zebrafish optic tec-tum. Neuron 76:1147–1160. CrossRef Medline

Gosse NJ, Nevin LM, Baier H (2008) Retinotopic order in the absence ofaxon competition. Nature 452:892– 895. CrossRef Medline

Hadjieconomou D, Rotkopf S, Alexandre C, Bell DM, Dickson BJ, Salecker I(2011) Flybow: genetic multicolor cell labeling for neural circuit analysisin Drosophila melanogaster. Nat Methods 8:260 –266. CrossRef Medline

Hampel S, Chung P, McKellar CE, Hall D, Looger LL, Simpson JH (2011)Drosophila Brainbow: a recombinase-based fluorescence labeling tech-nique to subdivide neural expression patterns. Nat Methods 8:253–259.CrossRef Medline

5038 • J. Neurosci., March 13, 2013 • 33(11):5027–5039 Robles et al. • Sublaminar Targeting of Retinal Axons

Hong YK, Kim IJ, Sanes JR (2011) Stereotyped axonal arbors of retinal gan-glion cell subsets in the mouse superior colliculus. J Comp Neurol 519:1691–1711. CrossRef Medline

Hua JY, Smear MC, Baier H, Smith SJ (2005) Regulation of axon growth invivo by activity-based competition. Nature 434:1022–1026. CrossRefMedline

Huberman AD, Manu M, Koch SM, Susman MW, Lutz AB, Ullian EM,Baccus SA, Barres BA (2008) Architecture and activity-mediated refine-ment of axonal projections from a mosaic of genetically identified retinalganglion cells. Neuron 59:425– 438. CrossRef Medline

Huberman AD, Wei W, Elstrott J, Stafford BK, Feller MB, Barres BA (2009)Genetic identification of an On-Off direction-selective retinal ganglioncell subtype reveals a layer-specific subcortical map of posterior motion.Neuron 62:327–334. CrossRef Medline

Huberman AD, Clandinin TR, Baier H (2010) Molecular and cellular mech-anisms of lamina-specific axon targeting. Cold Spring Harb Perspect Biol2:a001743. CrossRef Medline

Inoue A, Sanes JR (1997) Lamina-specific connectivity in the brain: regulationby N-cadherin, neurotrophins, and glycoconjugates. Science 276:1428–1431.CrossRef Medline

Kim IJ, Zhang Y, Meister M, Sanes JR (2010) Laminar restriction of retinalganglion cell dendrites and axons: subtype-specific developmental pat-terns revealed with transgenic markers. J Neurosci 30:1452–1462.CrossRef Medline

Le X, Langenau DM, Keefe MD, Kutok JL, Neuberg DS, Zon LI (2007) Heatshock-inducible Cre/Lox approaches to induce diverse types of tumorsand hyperplasia in transgenic zebrafish. Proc Natl Acad Sci U S A 104:9410 –9415. CrossRef Medline

Lee PH, Helms MC, Augustine GJ, Hall WC (1997) Role of intrinsic synap-tic circuitry in collicular sensorimotor integration. Proc Natl Acad SciU S A 94:13299 –13304. CrossRef Medline

Livet J, Weissman TA, Kang H, Draft RW, Lu J, Bennis RA, Sanes JR, Licht-man JW (2007) Transgenic strategies for combinatorial expression offluorescent proteins in the nervous system. Nature 450:56 – 62. CrossRefMedline

Luo L, Flanagan JG (2007) Development of continuous and discrete neuralmaps. Neuron 56:284 –300. CrossRef Medline

Mangrum WI, Dowling JE, Cohen ED (2002) A morphological classifica-tion of ganglion cells in the zebrafish retina. Vis Neurosci 19:767–779.CrossRef Medline

Masland RH (2001) The fundamental plan of the retina. Nat Neurosci4:877– 886. CrossRef Medline

May PJ (2006) The mammalian superior colliculus: laminar structure andconnections. Prog Brain Res 151:321–378. CrossRef Medline

McLaughlin T, O’Leary DD (2005) Molecular gradients and developmentof retinotopic maps. Annu Rev Neurosci 28:327–355. CrossRef Medline

McLaughlin T, Torborg CL, Feller MB, O’Leary DD (2003) Retinotopicmap refinement requires spontaneous retinal waves during a brief criticalperiod of development. Neuron 40:1147–1160. CrossRef Medline

Meek J (1983) Functional anatomy of the tectum mesencephali of the gold-fish. An explorative analysis of the functional implications of the laminarstructural organization of the tectum. Brain Res 287:247–297. Medline

Meyer MP, Smith SJ (2006) Evidence from in vivo imaging that synaptogen-esis guides the growth and branching of axonal arbors by two distinctmechanisms. J Neurosci 26:3604 –3614. CrossRef Medline

Mooney RD, Klein BG, Jacquin MF, Rhoades RW (1984) Dendrites of deeplayer, somatosensory superior collicular neurons extend into the superfi-cial laminae. Brain Res 324:361–365. CrossRef Medline

Mumm JS, Williams PR, Godinho L, Koerber A, Pittman AJ, Roeser T, ChienCB, Baier H, Wong RO (2006) In vivo imaging reveals dendritic target-ing of laminated afferents by zebrafish retinal ganglion cells. Neuron 52:609 – 621. CrossRef Medline

Nevin LM, Taylor MR, Baier H (2008) Hardwiring of fine synaptic layers inthe zebrafish visual pathway. Neural Dev 3:36. CrossRef Medline

Nevin LM, Robles E, Baier H, Scott EK (2010) Focusing on optic tectumcircuitry through the lens of genetics. BMC Biol 8:126. CrossRef Medline

Niell CM, Smith SJ (2005) Functional imaging reveals rapid development ofvisual response properties in the zebrafish tectum. Neuron 45:941–951.CrossRef Medline

Nikolaou N, Lowe AS, Walker AS, Abbas F, Hunter PR, Thompson ID, MeyerMP (2012) Parametric functional maps of visual inputs to the tectum.Neuron 76:317–324. CrossRef Medline

Ott M, Walz BC, Paulsen UJ, Mack AF, Wagner HJ (2007) Retinotectalganglion cells in the zebrafish, Danio rerio. J Comp Neurol 501:647– 658.CrossRef Medline

Pan YA, Livet J, Sanes JR, Lichtman JW, Schier AF (2011) Multicolor Brain-bow imaging in zebrafish. Cold Spring Harb Protoc 2011:pdb.prot5546.CrossRef Medline

Pittman AJ, Law MY, Chien CB (2008) Pathfinding in a large vertebrateaxon tract: isotypic interactions guide retinotectal axons at multiplechoice points. Development 135:2865–2871. CrossRef Medline

Pologruto TA, Sabatini BL, Svoboda K (2003) ScanImage: flexible softwarefor operating laser scanning microscopes. Biomed Eng Online 2:13.CrossRef Medline

Ramon y Cajal S (1995) Histology of the nervous system of man and verte-brates. New York: Oxford UP.

Robles E, Smith SJ, Baier H (2011) Characterization of genetically targetedneuron types in the zebrafish optic tectum. Front Neural Circuits 5:1.CrossRef Medline

Rockhill RL, Daly FJ, MacNeil MA, Brown SP, Masland RH (2002) The diver-sity of ganglion cells in a mammalian retina. J Neurosci 22:3831–3843.Medline

Sachs GM, Jacobson M, Caviness VS Jr (1986) Postnatal changes in ar-borization patterns of murine retinocollicular axons. J Comp Neurol 246:395– 408. CrossRef Medline

Scott EK, Baier H (2009) The cellular architecture of the larval zebrafishtectum, as revealed by gal4 enhancer trap lines. Front Neural Circuits3:13. CrossRef Medline

Simon DK, O’Leary DD (1992) Development of topographic order in themammalian retinocollicular projection. J Neurosci 12:1212–1232.Medline

Smear MC, Tao HW, Staub W, Orger MB, Gosse NJ, Liu Y, Takahashi K, PooMM, Baier H (2007) Vesicular glutamate transport at a central synapselimits the acuity of visual perception in zebrafish. Neuron 53:65–77.CrossRef Medline

Straw AD (2008) Vision egg: an open-source library for realtime visual stim-ulus generation. Front Neuroinform 2:4. CrossRef Medline

Vaney DI, Sivyer B, Taylor WR (2012) Direction selectivity in the retina:symmetry and asymmetry in structure and function. Nat Rev Neurosci13:194 –208. CrossRef Medline

Volgyi B, Chheda S, Bloomfield SA (2009) Tracer coupling patterns of theganglion cell subtypes in the mouse retina. J Comp Neurol 512:664 – 687.CrossRef Medline

Wang L, Sarnaik R, Rangarajan K, Liu X, Cang J (2010) Visual receptive fieldproperties of neurons in the superficial superior colliculus of the mouse.J Neurosci 30:16573–16584. CrossRef Medline

Weber T, Bohm G, Hermann E, Schutz G, Schonig K, Bartsch D (2009)Inducible gene manipulations in serotonergic neurons. Front Mol Neu-rosci 2:24. CrossRef Medline

Wei W, Feller MB (2011) Organization and development of direction-selective circuits in the retina. Trends Neurosci 34:638 – 645. CrossRefMedline

Xiao T, Baier H (2007) Lamina-specific axonal projections in the zebrafishtectum require the type IV collagen Dragnet. Nat Neurosci 10:1529 –1537.CrossRef Medline

Xiao T, Roeser T, Staub W, Baier H (2005) A GFP-based genetic screenreveals mutations that disrupt the architecture of the zebrafish retinotec-tal projection. Development 132:2955–2967. CrossRef Medline

Xiao T, Staub W, Robles E, Gosse NJ, Cole GJ, Baier H (2011) Assembly oflamina-specific neuronal connections by slit bound to type IV collagen.Cell 146:164 –176. CrossRef Medline

Robles et al. • Sublaminar Targeting of Retinal Axons J. Neurosci., March 13, 2013 • 33(11):5027–5039 • 5039